Simulation of a connectionist stereo algorithm on a shared-memory multiprocessor

This paper presents results on the simulation of a connectionist network for stereo matching on a shared-memory multiprocessor. The nodes in the network represent possible matches between points in each image. Because the authors only consider matches between intensity edges, only a few of these nodes actually represent candidate matches for a give pair of images; consequently, only the active part of the network is ever constructed by the simulation. This includes the nodes representing candidate matches and the connections between these candidate matches as defined by a variety of constraints. Thus, the simulation involves constructing a list of matches and a list of connections, and simulating the iterations of the network. Each of these phases can be partitioned among a number of processors with very little interference between the processors due to synchronization or mutual exclusion. The resulting parallel implementation produced nearly linear speed-ups for up to the nine processors in the system.